@code-yeongyu/senpi 2026.7.9-2 → 2026.7.9

package/node_modules/@earendil-works/pi-tui/package.json

@@ -1,7 +1,7 @@
 {
 	"name": "@earendil-works/pi-tui",
 	"private": true,
-	"version": "2026.7.9-2",
+	"version": "2026.7.9",
 	"description": "Terminal User Interface library with differential rendering for efficient text-based applications",
 	"type": "module",
 	"main": "dist/index.js",

package/npm-shrinkwrap.json

@@ -1,21 +1,21 @@
 {
 	"name": "@code-yeongyu/senpi",
-	"version": "2026.7.9-2",
+	"version": "2026.7.9",
 	"lockfileVersion": 3,
 	"requires": true,
 	"packages": {
 		"": {
 			"name": "@code-yeongyu/senpi",
-			"version": "2026.7.9-2",
+			"version": "2026.7.9",
 			"license": "MIT",
 			"dependencies": {
 				"@anthropic-ai/sdk": "0.91.1",
 				"@aws-sdk/client-bedrock-runtime": "3.1048.0",
-				"@code-yeongyu/senpi-codemode": "^2026.7.9-2",
-				"@earendil-works/pi-agent-core": "^2026.7.9-2",
-				"@earendil-works/pi-ai": "^2026.7.9-2",
-				"@earendil-works/pi-pty": "^2026.7.9-2",
-				"@earendil-works/pi-tui": "^2026.7.9-2",
+				"@code-yeongyu/senpi-codemode": "^2026.7.9",
+				"@earendil-works/pi-agent-core": "^2026.7.9",
+				"@earendil-works/pi-ai": "^2026.7.9",
+				"@earendil-works/pi-pty": "^2026.7.9",
+				"@earendil-works/pi-tui": "^2026.7.9",
 				"@mistralai/mistralai": "2.2.6",
 				"@modelcontextprotocol/sdk": "1.29.0",
 				"@mozilla/readability": "0.6.0",
@@ -554,11 +554,11 @@
 			}
 		},
 		"node_modules/@code-yeongyu/senpi-codemode": {
-			"version": "2026.7.9-2",
+			"version": "2026.7.9",
 			"inBundle": true,
 			"license": "MIT",
 			"dependencies": {
-				"@earendil-works/pi-ai": "^2026.7.9-2",
+				"@earendil-works/pi-ai": "^2026.7.9",
 				"typebox": "1.1.38"
 			},
 			"peerDependencies": {
@@ -703,11 +703,11 @@
 			]
 		},
 		"node_modules/@earendil-works/pi-agent-core": {
-			"version": "2026.7.9-2",
+			"version": "2026.7.9",
 			"inBundle": true,
 			"license": "MIT",
 			"dependencies": {
-				"@earendil-works/pi-ai": "^2026.7.9-2",
+				"@earendil-works/pi-ai": "^2026.7.9",
 				"ignore": "7.0.5",
 				"typebox": "1.1.38",
 				"yaml": "2.9.0"
@@ -717,7 +717,7 @@
 			}
 		},
 		"node_modules/@earendil-works/pi-ai": {
-			"version": "2026.7.9-2",
+			"version": "2026.7.9",
 			"inBundle": true,
 			"license": "MIT",
 			"dependencies": {
@@ -745,7 +745,7 @@
 			}
 		},
 		"node_modules/@earendil-works/pi-pty": {
-			"version": "2026.7.9-2",
+			"version": "2026.7.9",
 			"inBundle": true,
 			"license": "MIT",
 			"dependencies": {
@@ -756,7 +756,7 @@
 			}
 		},
 		"node_modules/@earendil-works/pi-tui": {
-			"version": "2026.7.9-2",
+			"version": "2026.7.9",
 			"inBundle": true,
 			"license": "MIT",
 			"dependencies": {

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@code-yeongyu/senpi",
-	"version": "2026.7.9-2",
+	"version": "2026.7.9",
 	"description": "Coding agent CLI with read, bash, edit, write tools and session management",
 	"type": "module",
 	"piConfig": {
@@ -44,11 +44,11 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "0.91.1",
 		"@aws-sdk/client-bedrock-runtime": "3.1048.0",
-		"@code-yeongyu/senpi-codemode": "^2026.7.9-2",
-		"@earendil-works/pi-agent-core": "^2026.7.9-2",
-		"@earendil-works/pi-ai": "^2026.7.9-2",
-		"@earendil-works/pi-pty": "^2026.7.9-2",
-		"@earendil-works/pi-tui": "^2026.7.9-2",
+		"@code-yeongyu/senpi-codemode": "^2026.7.9",
+		"@earendil-works/pi-agent-core": "^2026.7.9",
+		"@earendil-works/pi-ai": "^2026.7.9",
+		"@earendil-works/pi-pty": "^2026.7.9",
+		"@earendil-works/pi-tui": "^2026.7.9",
 		"@mistralai/mistralai": "2.2.6",
 		"@modelcontextprotocol/sdk": "1.29.0",
 		"@mozilla/readability": "0.6.0",

package/dist/core/cache-stats.d.ts

@@ -1,49 +0,0 @@
-import type { AssistantMessage } from "@earendil-works/pi-ai";
-import type { SessionEntry } from "./session-manager.ts";
-/**
- * Prompt-cache TTL: idle gaps longer than this are worth mentioning as the
- * likely cause of a miss. Anthropic's default cache TTL is 5 minutes.
- */
-export declare const CACHE_TTL_MS: number;
-/** A counted cache miss on a single assistant message. */
-export interface CacheMiss {
-    /** Prompt tokens that were in the previous turn's prompt but not read from cache. */
-    missedTokens: number;
-    /** Extra dollars paid vs. a full cache hit; 0 when pricing is unknown. */
-    missedCost: number;
-    /** Milliseconds since the previous request (which last refreshed the cache). */
-    idleMs: number;
-    /** True when the model changed relative to the previous request. */
-    modelChanged: boolean;
-}
-export interface CacheWasteTotals {
-    missedTokens: number;
-    missedCost: number;
-    /** Number of counted misses (turns above the noise floor). */
-    missCount: number;
-}
-/** Minimal pricing lookup, satisfied by ModelRegistry. Cost is $/million tokens. */
-export interface ModelPriceSource {
-    find(provider: string, modelId: string): {
-        cost: {
-            cacheRead: number;
-        };
-    } | undefined;
-}
-/**
- * Cumulative cache waste across a session: prompt tokens that should have been
- * cache reads (they were in the previous turn's prompt) but were re-billed.
- */
-export declare function computeCacheWaste(entries: SessionEntry[], models: ModelPriceSource): CacheWasteTotals;
-/**
- * All counted cache misses across a session, keyed by the assistant message
- * (by reference) that paid for them. Used to re-derive transcript notices when
- * rebuilding the chat from entries (resume, post-compaction rebuild).
- */
-export declare function collectCacheMisses(entries: SessionEntry[], models: ModelPriceSource): Map<AssistantMessage, CacheMiss>;
-/**
- * Detect a cache miss on a just-completed assistant message.
- * `entries` must not yet contain `message` (message_end fires before persistence).
- */
-export declare function detectCacheMiss(entries: SessionEntry[], message: AssistantMessage, models: ModelPriceSource): CacheMiss | undefined;
-//# sourceMappingURL=cache-stats.d.ts.map

package/dist/core/cache-stats.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"cache-stats.d.ts","sourceRoot":"","sources":["../../src/core/cache-stats.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,uBAAuB,CAAC;AAC9D,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,sBAAsB,CAAC;AAEzD;;;GAGG;AACH,eAAO,MAAM,YAAY,QAAgB,CAAC;AAK1C,0DAA0D;AAC1D,MAAM,WAAW,SAAS;IACzB,qFAAqF;IACrF,YAAY,EAAE,MAAM,CAAC;IACrB,0EAA0E;IAC1E,UAAU,EAAE,MAAM,CAAC;IACnB,gFAAgF;IAChF,MAAM,EAAE,MAAM,CAAC;IACf,oEAAoE;IACpE,YAAY,EAAE,OAAO,CAAC;CACtB;AAED,MAAM,WAAW,gBAAgB;IAChC,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;IACnB,8DAA8D;IAC9D,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,oFAAoF;AACpF,MAAM,WAAW,gBAAgB;IAChC,IAAI,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG;QAAE,IAAI,EAAE;YAAE,SAAS,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,GAAG,SAAS,CAAC;CACrF;AAmGD;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,gBAAgB,GAAG,gBAAgB,CAErG;AAED;;;;GAIG;AACH,wBAAgB,kBAAkB,CACjC,OAAO,EAAE,YAAY,EAAE,EACvB,MAAM,EAAE,gBAAgB,GACtB,GAAG,CAAC,gBAAgB,EAAE,SAAS,CAAC,CAElC;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAC9B,OAAO,EAAE,YAAY,EAAE,EACvB,OAAO,EAAE,gBAAgB,EACzB,MAAM,EAAE,gBAAgB,GACtB,SAAS,GAAG,SAAS,CAEvB"}

package/dist/core/cache-stats.js

@@ -1,101 +0,0 @@
-/**
- * Prompt-cache TTL: idle gaps longer than this are worth mentioning as the
- * likely cause of a miss. Anthropic's default cache TTL is 5 minutes.
- */
-export const CACHE_TTL_MS = 5 * 60 * 1000;
-/** Per-turn misses at or below this are cache breakpoint granularity noise. */
-const NOISE_FLOOR_TOKENS = 1024;
-/**
- * Compute the cache miss for one assistant message relative to the previous
- * request. Returns undefined when nothing is counted: first turn, after a
- * reset, no cache activity ever reported (provider without cache support), or
- * miss below the noise floor.
- */
-function detectMiss(prev, message, models) {
-    const usage = message.usage;
-    const promptTokens = usage.input + usage.cacheRead + usage.cacheWrite;
-    // A zero-cache turn only counts when cache activity was reported before:
-    // on cache-read-only providers that is a total miss, while on providers
-    // that never report caching it means nothing.
-    if (!prev || promptTokens <= 0 || (usage.cacheRead + usage.cacheWrite === 0 && !prev.reportedCache)) {
-        return undefined;
-    }
-    const missedTokens = Math.min(prev.promptTokens, promptTokens) - usage.cacheRead;
-    if (missedTokens <= NOISE_FLOOR_TOKENS)
-        return undefined;
-    // Extra cost = missed tokens billed at the actual paid rate (input/cacheWrite,
-    // incl. write premium) instead of the cache-read rate. Missed tokens can only
-    // land in the input or cacheWrite buckets, so the paid rate comes straight
-    // from this message's own cost breakdown.
-    const paidTokens = usage.input + usage.cacheWrite;
-    const paidPerToken = paidTokens > 0 ? (usage.cost.input + usage.cost.cacheWrite) / paidTokens : 0;
-    const readPerToken = usage.cacheRead > 0
-        ? usage.cost.cacheRead / usage.cacheRead
-        : (models.find(message.provider, message.model)?.cost.cacheRead ?? 0) / 1_000_000;
-    return {
-        missedTokens,
-        missedCost: missedTokens * Math.max(0, paidPerToken - readPerToken),
-        idleMs: Math.max(0, message.timestamp - prev.timestamp),
-        modelChanged: `${message.provider}/${message.model}` !== prev.modelKey,
-    };
-}
-function asPreviousRequest(message, reportedCache) {
-    const usage = message.usage;
-    const promptTokens = usage.input + usage.cacheRead + usage.cacheWrite;
-    if (promptTokens <= 0)
-        return undefined;
-    return {
-        promptTokens,
-        modelKey: `${message.provider}/${message.model}`,
-        timestamp: message.timestamp,
-        reportedCache: reportedCache || usage.cacheRead + usage.cacheWrite > 0,
-    };
-}
-function scan(entries, models) {
-    let prev;
-    const totals = { missedTokens: 0, missedCost: 0, missCount: 0 };
-    const misses = new Map();
-    for (const entry of entries) {
-        if (entry.type === "compaction" || entry.type === "branch_summary") {
-            // The context legitimately changed; the next turn's prompt is new content,
-            // not re-billed content. Model switches are NOT exempt: they re-bill the
-            // full prompt and should be counted.
-            prev = undefined;
-            continue;
-        }
-        if (entry.type === "message" && entry.message.role === "assistant") {
-            const miss = detectMiss(prev, entry.message, models);
-            if (miss) {
-                totals.missedTokens += miss.missedTokens;
-                totals.missedCost += miss.missedCost;
-                totals.missCount += 1;
-                misses.set(entry.message, miss);
-            }
-            prev = asPreviousRequest(entry.message, prev?.reportedCache ?? false) ?? prev;
-        }
-    }
-    return { prev, totals, misses };
-}
-/**
- * Cumulative cache waste across a session: prompt tokens that should have been
- * cache reads (they were in the previous turn's prompt) but were re-billed.
- */
-export function computeCacheWaste(entries, models) {
-    return scan(entries, models).totals;
-}
-/**
- * All counted cache misses across a session, keyed by the assistant message
- * (by reference) that paid for them. Used to re-derive transcript notices when
- * rebuilding the chat from entries (resume, post-compaction rebuild).
- */
-export function collectCacheMisses(entries, models) {
-    return scan(entries, models).misses;
-}
-/**
- * Detect a cache miss on a just-completed assistant message.
- * `entries` must not yet contain `message` (message_end fires before persistence).
- */
-export function detectCacheMiss(entries, message, models) {
-    return detectMiss(scan(entries, models).prev, message, models);
-}
-//# sourceMappingURL=cache-stats.js.map

package/dist/core/cache-stats.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"cache-stats.js","sourceRoot":"","sources":["../../src/core/cache-stats.ts"],"names":[],"mappings":"AAGA;;;GAGG;AACH,MAAM,CAAC,MAAM,YAAY,GAAG,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC;AAE1C,+EAA+E;AAC/E,MAAM,kBAAkB,GAAG,IAAI,CAAC;AAuChC;;;;;GAKG;AACH,SAAS,UAAU,CAClB,IAAiC,EACjC,OAAyB,EACzB,MAAwB;IAExB,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;IAC5B,MAAM,YAAY,GAAG,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC;IACtE,yEAAyE;IACzE,wEAAwE;IACxE,8CAA8C;IAC9C,IAAI,CAAC,IAAI,IAAI,YAAY,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,EAAE,CAAC;QACrG,OAAO,SAAS,CAAC;IAClB,CAAC;IAED,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,YAAY,EAAE,YAAY,CAAC,GAAG,KAAK,CAAC,SAAS,CAAC;IACjF,IAAI,YAAY,IAAI,kBAAkB;QAAE,OAAO,SAAS,CAAC;IAEzD,+EAA+E;IAC/E,8EAA8E;IAC9E,2EAA2E;IAC3E,0CAA0C;IAC1C,MAAM,UAAU,GAAG,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,UAAU,CAAC;IAClD,MAAM,YAAY,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;IAClG,MAAM,YAAY,GACjB,KAAK,CAAC,SAAS,GAAG,CAAC;QAClB,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC,SAAS;QACxC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,EAAE,OAAO,CAAC,KAAK,CAAC,EAAE,IAAI,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,SAAS,CAAC;IAEpF,OAAO;QACN,YAAY;QACZ,UAAU,EAAE,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,YAAY,GAAG,YAAY,CAAC;QACnE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC;QACvD,YAAY,EAAE,GAAG,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,KAAK,EAAE,KAAK,IAAI,CAAC,QAAQ;KACtE,CAAC;AACH,CAAC;AAED,SAAS,iBAAiB,CAAC,OAAyB,EAAE,aAAsB;IAC3E,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;IAC5B,MAAM,YAAY,GAAG,KAAK,CAAC,KAAK,GAAG,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC;IACtE,IAAI,YAAY,IAAI,CAAC;QAAE,OAAO,SAAS,CAAC;IACxC,OAAO;QACN,YAAY;QACZ,QAAQ,EAAE,GAAG,OAAO,CAAC,QAAQ,IAAI,OAAO,CAAC,KAAK,EAAE;QAChD,SAAS,EAAE,OAAO,CAAC,SAAS;QAC5B,aAAa,EAAE,aAAa,IAAI,KAAK,CAAC,SAAS,GAAG,KAAK,CAAC,UAAU,GAAG,CAAC;KACtE,CAAC;AACH,CAAC;AAED,SAAS,IAAI,CACZ,OAAuB,EACvB,MAAwB;IAExB,IAAI,IAAiC,CAAC;IACtC,MAAM,MAAM,GAAqB,EAAE,YAAY,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC;IAClF,MAAM,MAAM,GAAG,IAAI,GAAG,EAA+B,CAAC;IAEtD,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,IAAI,KAAK,YAAY,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,EAAE,CAAC;YACpE,2EAA2E;YAC3E,yEAAyE;YACzE,qCAAqC;YACrC,IAAI,GAAG,SAAS,CAAC;YACjB,SAAS;QACV,CAAC;QACD,IAAI,KAAK,CAAC,IAAI,KAAK,SAAS,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,KAAK,WAAW,EAAE,CAAC;YACpE,MAAM,IAAI,GAAG,UAAU,CAAC,IAAI,EAAE,KAAK,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC;YACrD,IAAI,IAAI,EAAE,CAAC;gBACV,MAAM,CAAC,YAAY,IAAI,IAAI,CAAC,YAAY,CAAC;gBACzC,MAAM,CAAC,UAAU,IAAI,IAAI,CAAC,UAAU,CAAC;gBACrC,MAAM,CAAC,SAAS,IAAI,CAAC,CAAC;gBACtB,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;YACjC,CAAC;YACD,IAAI,GAAG,iBAAiB,CAAC,KAAK,CAAC,OAAO,EAAE,IAAI,EAAE,aAAa,IAAI,KAAK,CAAC,IAAI,IAAI,CAAC;QAC/E,CAAC;IACF,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;AACjC,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,iBAAiB,CAAC,OAAuB,EAAE,MAAwB;IAClF,OAAO,IAAI,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,MAAM,CAAC;AACrC,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,kBAAkB,CACjC,OAAuB,EACvB,MAAwB;IAExB,OAAO,IAAI,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,MAAM,CAAC;AACrC,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAC9B,OAAuB,EACvB,OAAyB,EACzB,MAAwB;IAExB,OAAO,UAAU,CAAC,IAAI,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,IAAI,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;AAChE,CAAC","sourcesContent":["import type { AssistantMessage } from \"@earendil-works/pi-ai\";\nimport type { SessionEntry } from \"./session-manager.ts\";\n\n/**\n * Prompt-cache TTL: idle gaps longer than this are worth mentioning as the\n * likely cause of a miss. Anthropic's default cache TTL is 5 minutes.\n */\nexport const CACHE_TTL_MS = 5 * 60 * 1000;\n\n/** Per-turn misses at or below this are cache breakpoint granularity noise. */\nconst NOISE_FLOOR_TOKENS = 1024;\n\n/** A counted cache miss on a single assistant message. */\nexport interface CacheMiss {\n\t/** Prompt tokens that were in the previous turn's prompt but not read from cache. */\n\tmissedTokens: number;\n\t/** Extra dollars paid vs. a full cache hit; 0 when pricing is unknown. */\n\tmissedCost: number;\n\t/** Milliseconds since the previous request (which last refreshed the cache). */\n\tidleMs: number;\n\t/** True when the model changed relative to the previous request. */\n\tmodelChanged: boolean;\n}\n\nexport interface CacheWasteTotals {\n\tmissedTokens: number;\n\tmissedCost: number;\n\t/** Number of counted misses (turns above the noise floor). */\n\tmissCount: number;\n}\n\n/** Minimal pricing lookup, satisfied by ModelRegistry. Cost is $/million tokens. */\nexport interface ModelPriceSource {\n\tfind(provider: string, modelId: string): { cost: { cacheRead: number } } | undefined;\n}\n\n/** The last request seen by the scan; everything in its prompt should be cached. */\ninterface PreviousRequest {\n\tpromptTokens: number;\n\tmodelKey: string;\n\ttimestamp: number;\n\t/**\n\t * Sticky: some earlier request in this scan segment reported cache activity.\n\t * Distinguishes a total miss on a cache-read-only provider (OpenAI-style,\n\t * writes unreported) from a provider that never reports caching at all.\n\t */\n\treportedCache: boolean;\n}\n\n/**\n * Compute the cache miss for one assistant message relative to the previous\n * request. Returns undefined when nothing is counted: first turn, after a\n * reset, no cache activity ever reported (provider without cache support), or\n * miss below the noise floor.\n */\nfunction detectMiss(\n\tprev: PreviousRequest | undefined,\n\tmessage: AssistantMessage,\n\tmodels: ModelPriceSource,\n): CacheMiss | undefined {\n\tconst usage = message.usage;\n\tconst promptTokens = usage.input + usage.cacheRead + usage.cacheWrite;\n\t// A zero-cache turn only counts when cache activity was reported before:\n\t// on cache-read-only providers that is a total miss, while on providers\n\t// that never report caching it means nothing.\n\tif (!prev || promptTokens <= 0 || (usage.cacheRead + usage.cacheWrite === 0 && !prev.reportedCache)) {\n\t\treturn undefined;\n\t}\n\n\tconst missedTokens = Math.min(prev.promptTokens, promptTokens) - usage.cacheRead;\n\tif (missedTokens <= NOISE_FLOOR_TOKENS) return undefined;\n\n\t// Extra cost = missed tokens billed at the actual paid rate (input/cacheWrite,\n\t// incl. write premium) instead of the cache-read rate. Missed tokens can only\n\t// land in the input or cacheWrite buckets, so the paid rate comes straight\n\t// from this message's own cost breakdown.\n\tconst paidTokens = usage.input + usage.cacheWrite;\n\tconst paidPerToken = paidTokens > 0 ? (usage.cost.input + usage.cost.cacheWrite) / paidTokens : 0;\n\tconst readPerToken =\n\t\tusage.cacheRead > 0\n\t\t\t? usage.cost.cacheRead / usage.cacheRead\n\t\t\t: (models.find(message.provider, message.model)?.cost.cacheRead ?? 0) / 1_000_000;\n\n\treturn {\n\t\tmissedTokens,\n\t\tmissedCost: missedTokens * Math.max(0, paidPerToken - readPerToken),\n\t\tidleMs: Math.max(0, message.timestamp - prev.timestamp),\n\t\tmodelChanged: `${message.provider}/${message.model}` !== prev.modelKey,\n\t};\n}\n\nfunction asPreviousRequest(message: AssistantMessage, reportedCache: boolean): PreviousRequest | undefined {\n\tconst usage = message.usage;\n\tconst promptTokens = usage.input + usage.cacheRead + usage.cacheWrite;\n\tif (promptTokens <= 0) return undefined;\n\treturn {\n\t\tpromptTokens,\n\t\tmodelKey: `${message.provider}/${message.model}`,\n\t\ttimestamp: message.timestamp,\n\t\treportedCache: reportedCache || usage.cacheRead + usage.cacheWrite > 0,\n\t};\n}\n\nfunction scan(\n\tentries: SessionEntry[],\n\tmodels: ModelPriceSource,\n): { prev: PreviousRequest | undefined; totals: CacheWasteTotals; misses: Map<AssistantMessage, CacheMiss> } {\n\tlet prev: PreviousRequest | undefined;\n\tconst totals: CacheWasteTotals = { missedTokens: 0, missedCost: 0, missCount: 0 };\n\tconst misses = new Map<AssistantMessage, CacheMiss>();\n\n\tfor (const entry of entries) {\n\t\tif (entry.type === \"compaction\" || entry.type === \"branch_summary\") {\n\t\t\t// The context legitimately changed; the next turn's prompt is new content,\n\t\t\t// not re-billed content. Model switches are NOT exempt: they re-bill the\n\t\t\t// full prompt and should be counted.\n\t\t\tprev = undefined;\n\t\t\tcontinue;\n\t\t}\n\t\tif (entry.type === \"message\" && entry.message.role === \"assistant\") {\n\t\t\tconst miss = detectMiss(prev, entry.message, models);\n\t\t\tif (miss) {\n\t\t\t\ttotals.missedTokens += miss.missedTokens;\n\t\t\t\ttotals.missedCost += miss.missedCost;\n\t\t\t\ttotals.missCount += 1;\n\t\t\t\tmisses.set(entry.message, miss);\n\t\t\t}\n\t\t\tprev = asPreviousRequest(entry.message, prev?.reportedCache ?? false) ?? prev;\n\t\t}\n\t}\n\treturn { prev, totals, misses };\n}\n\n/**\n * Cumulative cache waste across a session: prompt tokens that should have been\n * cache reads (they were in the previous turn's prompt) but were re-billed.\n */\nexport function computeCacheWaste(entries: SessionEntry[], models: ModelPriceSource): CacheWasteTotals {\n\treturn scan(entries, models).totals;\n}\n\n/**\n * All counted cache misses across a session, keyed by the assistant message\n * (by reference) that paid for them. Used to re-derive transcript notices when\n * rebuilding the chat from entries (resume, post-compaction rebuild).\n */\nexport function collectCacheMisses(\n\tentries: SessionEntry[],\n\tmodels: ModelPriceSource,\n): Map<AssistantMessage, CacheMiss> {\n\treturn scan(entries, models).misses;\n}\n\n/**\n * Detect a cache miss on a just-completed assistant message.\n * `entries` must not yet contain `message` (message_end fires before persistence).\n */\nexport function detectCacheMiss(\n\tentries: SessionEntry[],\n\tmessage: AssistantMessage,\n\tmodels: ModelPriceSource,\n): CacheMiss | undefined {\n\treturn detectMiss(scan(entries, models).prev, message, models);\n}\n"]}